Research map / Adversarial robustness and security / Certified and empirical robustness
Boosting the Robustness-Accuracy Trade-off of SNNs by Robust Temporal Self-Ensemble
Jihang Wang, Dongcheng Zhao, Ruolin Chen, Qian Zhang, Yi Zeng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use RTE instead of TRADES | SNN adversarial training · CIFAR-100 (T=4) | clean accuracy (%): 56.97 -> 59.50 | Improved |
| use RTE instead of TRADES | SNN adversarial training · Tiny-ImageNet (T=4) | clean accuracy (%): 47.76 -> 50.46 | Improved |
| use RTE instead of TRADES | SNN adversarial training · SVHN (T=4) | clean accuracy (%): 94.00 -> 93.59 | Hurt |
| use RTE instead of AT | SNN adversarial training · CIFAR-100 (T=4) | APGD-CE robust accuracy (%): 13.08 -> 17.77 | Improved |
| combine RTE with SR (sparsity regularization) (ablation) | RTE · CIFAR-100 (T=4) | APGD-CE robust accuracy (%): 17.77 -> 20.18 | Improved |
| combine RTE with SR (ablation) | RTE · CIFAR-100 (T=4) | clean accuracy (%): 59.50 -> 58.10 | Hurt |
| use RTE+SR instead of AT+SR | AT+SR · CIFAR-10 (T=4) | worst-case robust accuracy: -0.90 (text) | Hurt |
| use RTE+SR instead of AT+SR | AT+SR · CIFAR-10 (T=4) | clean accuracy (%): 78.55 -> 80.85 | Improved |
| RTE+SR vs strongest baseline on trade-off metric (clean+robust) | baselines · CIFAR-100 and Tiny-ImageNet | clean + robust accuracy: +3.92 and +4.80 points | Improved |
| increase timesteps T from 4 to 12 (ablation) | AT, RAT, TRADES, RTE · CIFAR-10/100 | robustness-accuracy trade-off | Improved |
| vary regularization gamma: RTE vs TRADES (ablation) | RTE · CIFAR-10/100 | trade-off (clean+robust) | Improved |
| add RTE vs AT/RAT/TRADES in transferability matrices | RTE · CIFAR-10 T=8 | KL transferability | Improved |
Closest papers by meaning
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.92)
- Synergy Between the Strong and the Weak: Spiking Neural Networks are Inherently Self-Distillers (NeurIPS 2025 · similarity 0.90)
- Rethinking Spiking Neural Networks from an Ensemble Learning Perspective (ICLR 2025 · similarity 0.90)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.90)
- Robust Stable Spiking Neural Networks (ICML 2024 · similarity 0.89)
- Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
